Statistical Analysis Of Next Generation Sequencing Data PDF Download

The advent of next-generation sequencing (NGS) technologies has significantly advanced sequence-based genomic research and biomedical applications. Although a wide range of statistical methods and tools have been subsequently developed to support the analysis of NGS data in different steps and aspects, challenges continue to arise due to multiple issues. The central theme of this dissertation is to address the challenges and issues in three aspects of NGS analyses: sequencing alignment, Single Nucleotide Polymorphism (SNP) detection, and differential methylation identification. First, to investigate issues of low sequencing quality and repetitive reads in alignment, four commonly used alignment algorithms (SOAP2, Bowtie, BWA, and Novoalign) have been thoroughly reviewed and evaluated. The results show that the concordance among the algorithms is relatively low in reads with low sequencing quality, but can be substantially improved by trimming off low quality bases before alignment. As for aligning reads from repetitive regions, the simulation analysis shows that reads from repetitive regions tend to be aligned incorrectly, and suppressing reads with multiple hits can improve alignment accuracy significantly. Second, to address the challenges in SNP detection caused by low coverage, four SNP calling algorithms (SOAPsnp, Atlas-SNP2, SAMtools, and GATK) have been compared and evaluated in a low-coverage single-sample sequencing dataset. Although the four algorithms have low agreement, GATK and Atlas-SNP2 show relatively higher calling rates and sensitivity than others programs. Third, a new hidden Markov model-based approach, HMM-DM, has been developed to identify differentially methylated regions (DMRs) in bisulfite sequencing data. This method well accounts for the large within group variation of methylation levels and can detect differential methylation in single-base resolution. It has been demonstrated to have superior performance compared with BSmooth, and its application has been illustrated using a real sequencing dataset. In the last part of this thesis, five DMR identification methods (methylKit, BSmooth, BiSeq, HMM-DM, and HMM-Fisher) have been systematically reviewed and compared using bisulfite sequencing datasets. All five methods show higher accuracy in the identification of simulated DMRs that are relatively long and have small within group variation. Compared with the three other methods, HMM-DM and HMM-Fisher yield relatively higher sensitivity and lower false positive rates, especially in DMRs with large within group variation. However, in the real data analysis, the five methods show low concordances, probably due to the different approaches they are taking when tackling the issues in DMR identification. Therefore, to guarantee a higher accuracy in validation and further analysis, users may choose the identified DMRs that are long and have small within group variation as a priority. In summary, this thesis has addressed several important questions in NGS studies through the development of new statistical methods and comprehensive bioinformatic analyses.